You have a 4.9. You have hundreds of reviews. And when a patient asks ChatGPT for the best injector in your city, you are not in the answer. Our read: these systems are built to summarize text, so the words written about you carry the answer more than the star average, and most practices with excellent ratings have reviews that say almost nothing a model can quote.
Reputation used to be a number you protected. Now it is a body of text that gets summarized, quoted and compared on your behalf, often before a patient ever sees your website. The practices winning that summary are not the ones with the highest rating. They are the ones whose patients, directories and treatment pages all describe the same specific work in the same specific language.
What an AI assistant actually reads when a patient asks for the best injector in your city
The surface has already moved. Research from Whitespark, published May 12, 2025, across 540 queries in Houston, Phoenix and Denver covering six local industries, found AI Overviews appearing on an average of 68 percent of local business-type queries, while local packs appeared on 39 percent. Aesthetics was not one of the industries studied, so read it as local business queries rather than patient queries. Split by intent, it gets sharper. Classic local queries like "med spa Scottsdale" still return a local pack 93 percent of the time, and AI Overviews appear on only 15 percent of local-intent queries. But informational queries, the ones patients ask first, like how long a filler result lasts, return an AI Overview 92 percent of the time and a local pack just 6 percent. Hybrid queries return an AI Overview 97 percent of the time.
That matters because aesthetic patients do not start with a name. They start with a question. Months of quiet research happen in that informational layer, and the layer is now answered by a summary rather than a list of businesses.
What feeds the summary is broader than your website. In Whitespark's sample, 60 percent of AI Overview citations pointed to third-party publishers and 40 percent to individual business sites. So when a model assembles an answer about your city, it is reading:
- The full text of your reviews, not the average
- Your Google Business Profile categories, services and Q&A
- Your treatment pages and the language on them
- Directory listings and whether they agree with each other
- Third-party mentions, editorial coverage and forum discussion
The patient side has moved too. A RepuGen survey of 619 patients, run by a reputation vendor and worth reading as directional, found 51 percent had adopted new internet-based tools or behaviors when researching providers, such as AI tools, review sites or other online search methods. The AI-specific number is 39.7 percent, who now use either AI tools or a mix of AI and traditional search. Among the 246 AI users in the sample, AI accounts for 56.8 percent of their searching, and 55.3 percent said they would trust AI to rank doctors.
In aesthetics specifically, The Medical Aesthetics AI Visibility Index 2026, released April 27 by the PR firm 5WPR with Haute MD and covered by AmSpa on May 6, 2026, analyzed more than 60 patient-intent queries across ChatGPT, Claude, Perplexity and Google AI Overviews to rank 25 leading brands by AI citation share. Like the RepuGen survey, it is run by an interested party, so read it as directional. It describes AI-generated responses increasingly functioning as the first consultation, influencing which brands and treatments patients consider before contacting a provider. It also reports that some widely used brands rank lower on citation share than their market presence would suggest. One caveat: the index ranks product brands such as Botox, Juvederm, CoolSculpting, SkinCeuticals and Morpheus8, not practices, though it notes some AI citations reference individual dermatology and plastic surgery practices. Excellent clinical work that is not written down anywhere does not exist to a model.
Why a 4.9 made of one-word reviews loses to a 4.7 that names the treatment
Here is the trade in plain terms. A model asked to recommend an injector needs to produce a sentence a patient finds useful. It cannot build that sentence from "Great!" and "Love this place." It can build it from "Dr. Patel did my lip filler, kept it conservative like I asked, and the swelling was gone in four days."
One of those reviews is retrievable. The other is decoration.
That is the argument for detailed, treatment-named, provider-named patient language over a higher average built on one-word praise. The higher average wins the badge. The detailed reviews supply the specifics, the treatment vocabulary and the named provider that let a model say something concrete. We have not seen a study that puts a number on the trade, so treat it as reasoning about how these systems retrieve text, not as a measured finding.
How to ask for reviews that produce real language
The instinct is to script patients. Do not. Scripted reviews read identically, carry obvious risk, and produce exactly the generic text you are trying to escape.
Ask better questions instead. Replace "please leave us a review" with a prompt that makes specificity the easy path:
- "If you have a minute, would you mention which treatment you had and how the first week went?"
- "What would you tell a friend who was nervous about this?"
- "Who looked after you, and what did they do that helped?"
Three changes in operations do most of the work. Ask at the right moment, which for aesthetics is usually after the result has settled and the patient is happy with it rather than on the way out the door. Ask by name, because the staff member who performed the treatment gets better answers than a generic SMS. And automate the sequence rather than the wording, so the timing is reliable and the voice stays the patient's own.
Patients write about what you ask them about. If nobody asks about the treatment, nobody names the treatment.
The consistency checklist: name every treatment the same way, everywhere
Patients and models both speak in treatment and brand names. A practice whose site says "injectables," whose profile says "skin care," and whose directory listing says "medical spa" has no way to match a query about a specific treatment in a specific city.
Work through this:
- Name treatments individually. One page per treatment, named the way patients say it, not bundled under a category heading.
- List services separately in your Google Business Profile. Each treatment as its own service entry, with a description that matches the language on your site.
- Choose categories deliberately. Primary and secondary categories should reflect what you actually want to be found for, not the broadest label available.
- Make directories agree. Same practice name, same address format, same phone, same service language. Disagreement across sources is the fastest way to get left out of a summary.
- Keep photos and Q&A current. Q&A in particular is readable text that answers the exact informational questions that now resolve in AI.
- Write the informational layer. The questions patients ask before they ask for a name are the queries most likely to produce an AI answer. If you have not written the honest answer to "how long does this last" and "what is recovery actually like," someone else has.
This is the Trust Stack in practice. Trust is built in layers and patients commit when every layer reinforces the next. Miss one and confidence breaks down, for a patient reading and for a model summarizing.
Replying to a negative review when you cannot discuss the patient
Generic reputation advice says to reply with specifics. In a clinical setting that advice is dangerous. The safe and effective reply does three things: it acknowledges a general service concern, it states how the practice handles that kind of concern, and it moves the conversation to a private channel.
What a public reply should not do is confirm that the reviewer is a patient, or reference any treatment, visit, record or clinical detail. In December 2022, HHS Office for Civil Rights settled with New Vision Dental over impermissible disclosure of protected health information in responses to online reviews, involving patient names, treatment and insurance information. The practice agreed to a corrective action plan, and OCR's director stated that providers cannot disclose protected health information when responding to negative online reviews. The restriction is generally understood to hold even when the reviewer has disclosed details themselves. Requirements and enforcement are set by federal health privacy law and your state board, so have your own counsel review your standard response language before you deploy it.
Written that way, the reply still carries weight. "We ask every patient to contact our clinical lead directly within 48 hours if a result is not settling as expected, and we will always see you back" is useful context about how the practice operates, legible to a prospective patient and to a model, and it discloses nothing.
The monthly check: three prompts, two assistants, one log
You cannot manage what you never look at. Once a month, open ChatGPT and Gemini and run the three prompts a patient in your market would actually type:
- "Best [treatment] in [your city]"
- "Who should I see for [treatment] near [neighborhood]?"
- "[Your practice name] reviews, is it any good?"
Record three things each time: whether you appear, what is said about you, and which sources were cited. That log becomes your visibility baseline. Watch the cited sources especially, because that tells you where the answer is being assembled from, and most of it will not be your website.
Ratings were a scoreboard. The words are the asset now, and they take months to accumulate, which means the practices that start this quarter own the answer by next year.
We rebuilt one aesthetic brand's visibility for AI search platforms and their organic presence shifted entirely, surfacing in ChatGPT, Gemini and AI-assisted results rather than Google alone. If your clinical work is excellent and the calendar still is not full, the gap is usually not your rating. Book a free 15 minute discovery call and we will run your prompts with you and show you what the assistants are saying.
Frequently asked questions
Does my star rating still matter if AI reads review text?
Yes, but as a threshold rather than a differentiator. Below a certain rating you get filtered out of consideration entirely. Above it, the number stops separating you from competitors and the specificity of the review text takes over, which is why a 4.7 with detailed, treatment-named reviews can out-recommend a 4.9 built on one-word praise.
Can I tell patients what to write in their review?
Scripting patients produces identical-sounding reviews, carries obvious risk, and gives you exactly the generic text you are trying to escape. Ask better questions instead, such as which treatment they had and how the first week went. Automate the timing of the request, never the wording.
How do I respond to a negative review without breaking patient privacy?
Acknowledge the general service concern, state how your practice handles that type of concern, and move the conversation private. Never confirm the reviewer is a patient or reference any treatment, visit or record, even if the reviewer disclosed it first. Privacy requirements are set by federal health privacy law and your state board, so have counsel review your standard reply language.
Is my website enough, or do directories matter for AI answers?
Your site is well under half the input. In Whitespark's May 2025 research, 60 percent of AI Overview citations pointed to third-party publishers and 40 percent to individual business sites. Directory consistency, third-party mentions and editorial coverage carry more of the answer than most practices expect.
How often should I check whether AI assistants recommend my practice?
Monthly is enough. Run three prompts a real patient would type across ChatGPT and Gemini, covering best treatment in your city, who to see near a neighborhood, and your practice name plus reviews. Log whether you appear, what is said, and which sources were cited.
